Comparison of cluster detection using patients and events of medically treated self-inflicted injuries in Alberta, Canada

Abstract

Geographic regions may be routinely evaluated for high rates of disease through statistical cluster detection tests. Areas
with statistically higher rates can be targeted for further epidemiological study and possible interventions. Generally these
tests are performed on incident or prevalent cases of a disease, and are not necessarily applicable to the health services
context. We examine a new method appropriate for health services data and introduce this new method to health services researchers
through a case study of self-inflicted injury presentations to emergency departments (EDs). We consider spatial clustering
with two approaches. A case-based method uses only the number of patients whereas an event-based method uses the number of
patients and the number of ED presentations. Analyses are adjusted by the age and gender distribution and restricted to a
sub-group of the population. Several potential clusters are identified within each method. The case-based method identifies
10 geographic areas with statistically higher numbers of individuals seeking ED care and the event-based method identifies
nine geographic areas with statistically higher numbers of ED presentations. These potential clusters are not likely to have
occurred by chance alone, given the age and gender distribution. The two testing procedures illustrate different aspects of
the data: individuals and presentations. The potential clusters based on individual cases suggest areas that may need special
intervention programs and services. The clustering of presentations may represent areas that require increased provision of
health services.